[ Front page ] [ The edge ] [ Finished ] [ About ] [ Links ]
2.9 per cent
Held by GPT-6 Astra. Set 2026-09-03, and has stood for 26 days. Written down here 2026-09-07.
The Erdos problems are questions the mathematician Paul Erdos asked and could not answer, some of them left standing for fifty years. They are not exercises with a solution hidden from the model. Nobody knows the answers. A set of them has been turned into a test, and the best model in the world currently solves fewer than three in a hundred. That makes this the most useful number on this site, because almost every other test a model sits has an answer somebody already wrote down, and doing well on those demonstrates recovery rather than discovery. This one cannot be passed by remembering. It is the difference between a machine that can return what is known and a machine that can add to it, and on this measure the second thing has barely started.
That description was written when this frontier was opened, on 2026-09-07, and is never rewritten. Where it and the number above it disagree, the number is the reading and the description is history.
Where to check it: https://epoch.ai/data/ai-benchmarking-dashboard
The program reads it from https://epoch.ai/data/benchmarks.csv, which is the same figure in a form a script can parse.
Last checked: 2026-09-28
Previous: The longest task an AI can finish on its own Next: Unsolved integer sequences worked out
[ Front page ] [ The edge ] [ Finished ] [ About ] [ Links ]